🚀 go-pugleaf

A while ago, when John first posted that link, I read the
article.  One thing I found puzzling, was the formula used
implied a larger data packet would worsen the situation,
even though it would require fewer acks.  I posted a note
here to John about that, but he didn't reply to my question
on it.  If the data packets were, for example 20 times the
size of the acks, you should theoretically be fine with a
20:1 down/up ratio.

"Randy S." wrote:
> 
> > >So the only way a DSL line will be subject to an overloaded
> > >upstream channel is if the provider uses too high an asymmetry ratio (I think
> > >we calculated that the effect would be seen starting at around a 10:1 ratio,
> > >and be very serious at around 15:1).
> >
> > High ratios of downlink to uplink decrease (not increase) the risk of
> > saturation.  I presume you actually mean "oversubscription rate" (rather
> > than "asymmetry ratio").  Higher oversubscription rates would of course
> > increase the risk of uplink saturation, but downlink saturation would
> > probably occur first.  BTW, I understand that "oversubscription rates"
> > for consumer DSL are commonly on the order of 100:1.
> 
> Oops, now I'm lost again.  From what I understand, the uplink saturation happens
> (or CAN happen) when a large imbalance between uplink and downlink bandwidth
> exist.  I.e. your download speed is fast enough that the return ACKs will overload
> the bandwidth of the much smaller uplink.  So when I say 10:1 asymmetry ratio, I
> mean that the download bandwidth is 10 times the rate of the upload bandwidth.  In
> back of the envelope calculations I seemed to see that when downloading at the
> maximum rate and with a normal TCP receive window (say 32 or 64K), a connection
> with an asymmetry ratio of 10:1 would start to have delayed downstream packets due
> to late ACKs because the upstream channel couldn't send them fast enough.  An
> asymmetry ratio of 15:1 would drag you to a standstill.  This is obviously not the
> ONLY thing that can saturate the upstream link, but it would seem to be the main
> danger for uplink saturation for DSL.  Most DSL asymmetry ratios would seem to not
> be high enough to see this problem (1.472 Mbps down/256 Kbps which is common is a
> 5.75 ratio), but I've seen offers of DSL at rates as asymmetrical as 640 Kbps down
> / 64 Kbps up which is in danger territory.  Cable, as you point out, can saturate
> the upstream link in other ways.  Also as you point out, upstream of the DSLAM, DSL
> would seem likely to saturate it's downstream link well before it ever saturated
> it's upstream link .
> 
> > >Conversely, since cable aggregates the
> > >channels along each segment, it can overload the upstream channel even IF each
> > >individual asymmetry ratio is not too high.
> >
> > The shared nature of cable is the killer because (1) the aggregate
> > uplink capacity is typically much less then the aggregate downlink
> > capacity [unlike DSL, where the aggregate is usually symmetrical] and
> > (2) saturation of the shared uplink kills the downlink for everyone on
> > the segment.
> 
> Yes, this was the point you made that really caught my attention, and made LOTS of
> sense.
> 
> Randy S.


-- 
Replies sent via e-mail to this address will be promptly
ignored.
To reply, replace everything to the left of "@" with
"james.knott".